Author: Zhang Yifan
Editor: Shen Siqi
Source: Hard AI
After the focus on AI servers, the market's attention has shifted to AI smartphones and AI PCs.
With the upcoming launch of Qualcomm's Snapdragon 8 Gen4 based on the new ARM architecture, the market's interest in AI PCs is increasing. In this week's earnings call, Lenovo revealed some new developments about itself and the industry.
<h2>1. Definition of &#34;Five Elements&#34; for AI PCs</h2>
There are different definitions of AI PCs in the market. Currently, Microsoft's definition is widely recognized, including:
<blockquote>
<ol>
<li>
In terms of memory, basic AI models require 16GB of memory, standard AI models require 32GB, and advanced AI models require 64GB or more;
</li>
<li>
In terms of computing power, at least 40TOPS of computing power is required;
</li>
</ol>
</blockquote>
Lenovo has further refined its definition of AI PCs based on this, including five key elements:
<blockquote>
<ol>
<li>
Built-in compressed large models;
</li>
<li>
Equipped with an AI assistant with natural language interaction capabilities;
</li>
<li>
In terms of hardware, it needs to have CPU, GPU, and NPU simultaneously;
</li>
<li>
Can help users build personal knowledge bases;
</li>
<li>
Emphasizes the protection of users' personal privacy;
</li>
</ol>
</blockquote>
Combining Microsoft's AI PC memory and computing power standards, the form of AI PCs becomes clearer: AI PCs are not only intelligent and powerful, but also provide users with a more personalized and secure experience.
<h2>2. Predicted Changes in AI PC Penetration Rate</h2>
According to Lenovo's own definition, there are currently few products on the market that meet the requirements. Lenovo estimates that the proportion this year will be below ten percentage points.
Lenovo predicts that by 2024, approximately 20% of the PCs shipped by the company will be equipped with NPUs (Neural Processing Units). By 2026, according to Lenovo's strict definition, the penetration rate of AI PCs may reach 50% to 60%. Currently, Lenovo's annual shipments range from 55 to 60 million units, which means that around 36 million units will be AI PCs.
American marketing expert Geoffrey Moore has a &#34;Law of Accelerating Market Penetration,&#34; where he believes that when the market penetration rate of a new product or service reaches 15%, its growth rate will accelerate and eventually reach around 50%. After that, the growth rate will gradually slow down until it stabilizes.
Therefore, the shipment volume of AI PCs may gradually increase in 2024, and from 2025 to 2026, it will enter a phase of rapid growth, bringing new opportunities to the entire industry chain.
<h2>3. Market Share of x86 and ARM</h2>
The replacement of the x86 architecture by ARM is a core issue that the market has been focusing on recently, which will play an important role in the proportion of the AI era's device replacement trend.
During the meeting, Lenovo stated that there will be a coexistence of x86 and ARM architectures in the future
x86 instructions are usually long, often containing multiple tasks in a single long instruction. In contrast, ARM excels at breaking down a long instruction into multiple short instructions. In other words, ARM is good at decomposing complex tasks into simple tasks. This approach helps reduce CPU power consumption while maintaining high performance.
Intel and AMD have long dominated the market share in the x86 architecture era. However, with the increasing demand for power efficiency in the AI computing era, the market is gradually shifting towards the use of ARM architecture. NVIDIA, AMD, Qualcomm, and MediaTek have all introduced CPU processors based on ARM architecture.
However, Lenovo also mentioned that for AI processors, besides power consumption, there are many other important metrics to consider. In the future, there will definitely be fierce competition between x86 and ARM. Such competition is good news for AI server manufacturers, as it provides an advantage in terms of both cost and processor selection.
<h2>4. Changes in the Supply Chain</h2>
According to data released by Goldman Sachs, the lead time for GPU supply is continuously decreasing. NVIDIA also stated in this week's earnings conference that the supply shortage situation for the H100 graphics card has eased.
Lenovo provided more detailed data at this meeting: last year, the waiting period for GPUs was approximately 52 weeks. At the beginning of this year, this period had been reduced from 52 weeks to 26 weeks, and so far, the waiting period has further decreased.
The shortened supply cycle is good news for server suppliers such as Lenovo, Dell, HP, and Supermicro, as it means that server shipments will no longer be constrained by supply shortages. However, Lenovo also pointed out that it is still uncertain when the supply chain issues can be fully resolved.
<h2>5. Copilot and Xiaotian</h2>
Due to well-known reasons, domestic Windows computers cannot directly use Microsoft's Copilot. In order to address the issue of AI assistants on PCs, Lenovo has launched its own on-device large model assistant &#34;Lenovo Xiaotian&#34;.
Unlike Copilot, which relies on cloud operation, &#34;Lenovo Xiaotian&#34; can run entirely locally.
According to tech blogger &#34;Kazuk&#34;, &#34;Lenovo Xiaotian&#34; has both local and cloud operation modes, allowing users to freely switch based on their needs.
This design not only ensures user data privacy but also enables the utilization of cloud processing for complex tasks when needed.
In terms of power consumption, &#34;Kazuk's&#34; data shows that although the local model runs on the CPU, power consumption is very low. This large model runs in the background and requires approximately 4GB of memory. Additionally, the impact on the CPU during inference can be considered negligible.
In terms of functionality, the local model can perform tasks such as drafting emails, speech transcription, meeting summaries, etc., with performance equivalent to a GPT-3.5 level model, which is sufficient to meet daily office needs
Overall, Lenovo's conference call reveals an optimistic attitude towards AI PCs. In the era of ARM, competition among upstream chip manufacturers will intensify further. Lenovo stated that it is currently unclear who can quickly take the lead, but believes that intense competition upstream is beneficial for downstream PC manufacturers. Faster technological advancements are advantageous for the arrival of the replacement wave

LNVGY

80992

00992

By 2024, the volume will gradually increase, entering an acceleration phase from 2025 to 2026, with the penetration rate set to exceed 15% at an accelerating inflection point

What issues are being discussed in the AI PC era from the Lenovo conference call? | AI Dehydration